autoresearch

Automated code improvement works by running a defined metric command to establish a baseline, testing a hypothesis, implementing changes, and measuring results. The cycle repeats automatically until the target metric is reached or maximum iterations are exhausted.

You need a runnable metric command to measure results, git for branching and history, and the ability to create and switch branches per iteration. These prerequisites allow the workflow to isolate experiments and commit changes with descriptive messages.

The automated experimentation workflow handles stalled progress by terminating the loop when a metric reaches the target or maximum iterations are exhausted. It includes stall handling to prevent infinite loops when improvements plateau during the cycle.

Yes, automated code improvement works across languages and projects. It applies a repeatable workflow from baseline through analysis, hypothesis, implementation, measurement, and decision, requiring only a runnable metric command and git for version control.

The best way to measure and commit code changes against a target is to use an automated experiment cycle that establishes a baseline, implements improvements, measures results, and commits changes to git branches with descriptive messages until the target is met.